Skip to main content
An agent isn’t limited to what it can do on its own. When another agent already handles part of a job well, an agent can find and call that agent — hand it a self-contained task, let it work, and use the result to keep going. Like sub-agents, this happens automatically. Dataleap decides when calling another agent is the right move; there’s nothing to set up or switch on. This page explains the concept so you recognize it when you see it in a run.

What it looks like

Diagram: a requesting agent sends a self-contained task to a called agent, which runs on its own and returns a result

One agent calls another it has access to, then continues with the result.

The requesting agent sends a complete, standalone task. The called agent runs on its own and returns just the answer — so instead of rebuilding logic another agent already has, your agent reuses it. Dataleap typically reaches for this when a job spans specialized agents that each own one part, or when a recurring process should run the same way every time. You can see it in the Runs trace: a Called agent step names the agent that was reached, so you can follow the hand-off between them.

Who an agent can reach

An agent only ever reaches agents you already have access to:
  • Team agents reach other agents in the same team.
  • Personal agents reach your own agents, plus agents in every team you belong to.
Nothing crosses a boundary you’re not already on. For how teams and shared agents work, see team spaces.

Different from sub-agents

Sub-agents are private helpers an agent builds for itself, and they stay inside that agent’s Tools tab. Agent-to-agent communication is broader: an agent reaches independent agents that you or your teammates already built, and those calls appear in the Runs trace as separate agents — not tucked inside one agent’s tools.

Sub-agents & model tiers

The other way an agent leans on another — private helpers it builds for itself.

Team spaces

How teams and shared agents shape which agents an agent can reach.